regarding the respsize draft: preferring glue of certain types

[Paul Vixie <paul@vix.com>]

i may have a conflict during today's meeting in which case i'd like to
present my recent thoughts about respsize here and hope that kato or
others will represent them in real time.  these issues are highly topical
since they will drive the policies by which IANA will add, or not add,
AAAA RRs for root and tld servers.

---

in the event where a delegation response is going to be truncated, it
is desireable that glue of the appropriate type be added preferentially.
if the query arrived by IPv4, then A RRs ought to be added first; if
the query arrived by IPv6, then AAAA RRs ought to be added first.

this assumes that a dns client will wish to use the same protocol for
reaching the designated servers as it used to reach the delegating
server.

to prevent pathological starvation, at most one glue rrset of each type
should be present even in a truncated response, unless the rrsets
are such that the smallest rrset of each glue type would not fit.

since truncation must be on rrset boundaries, smaller rrsets should be
added before larger ones.  if the smallest rrset of any type is still
too large to fit, then the delegation response's additional data section
will be empty.

---

(before anybody suggests that we add a "glue order" option in EDNS, keep
in mind that when EDNS is available, truncation will only rarely occur,
even with both kinds of glue records and full dns security signatures.
this entire discussion has to do with imputing the best interests of dns
clients who lack the ability to express either preference or capability.)

#----------------------------------------------------------------------
# To unsubscribe, send a message to <dnsop-request@cafax.se>.